[NCLUG] RAID array not started on re-boot

[Kevin Olson]

I will try the suggestion of the mdadm --zero-superblock on the /dev/sda in
the near future, perhaps over the weekend if the power stays up. The
building has a power issue, and the input from the grid isn't working, so
everything is running on a generator, and then over the past weekend the
generator ran out of fuel. So, power predictability is a bit low at the
moment....

In the interim, I changed the /etc/mdadm.conf file to the following:

DEVICE /dev/sda1 /dev/sdb1 /dev/sde1 /dev/sdf1

ARRAY /dev/md2 metadata=1.2 name=dta.example.com:2 UUID=YYY
ARRAY /dev/md1 metadata=1.2 name=dta.example.com:1 UID=ZZZ
devices=/dev/sda1,/dev/sdb1

We had to shutdown the machine last night while they did some power work
and re-started it this morning. The RAID came up correctly this morning, so
I am hopeful the modification to the mdadm.conf will hold us for a while.
We are moving off this machine in couple of months, so as long as the array
starts, I think we will be OK.

Thank you for all of the helpful suggestions in debugging the issue!

Kevin


On Mon, Sep 9, 2013 at 4:54 PM, Jesse Griffin <jesse at tummy.com> wrote:

> You could possibly get away with running:
>
>       mdadm --zero-superblock /dev/sda
>
> After that, ensure that the RAID still looks good.  If there are problems
> then
> you could fail sda1 out of the RAID and then add it back in, something
> like:
>
>     mdadm --manage --fail /dev/md1 /dev/sda1
>     mdadm --manage --remove /dev/md1 /dev/sda1
>     mdadm --manage --add /dev/md1 /dev/sda1
>
> Of course, that process will initiate a resync from the data on sdb1.
>  Also, if
> you run into problems when re-adding sda1, then you can run the
> --zero-superblock command above on sda1, then try adding it again.
>
> Essentially, the above should give you a clean slate on sda/sda1, but may
> come
> at the cost of rebuilding the array.  If the process works as expected,
> then you
> could do the same steps on sdb/sdb1.
>
> Note: I'm not guaranteeing the above by any means, but those are some
> ideas to
> try if you are interested.
>
>
> Thank you,
> Jesse Griffin
> tummy.com, ltd.
>
>
>